Spatio-temporal variations in protistan communities along an O2/H2S gradient in the anoxic Framvaren Fjord (Norway)


  • Editor: Alfons Stams

Correspondence: Anke Behnke, Department of Biology, University of Kaiserslautern, Erwin Schroedinger Str. 14, D-67663 Kaiserslautern, Germany. Tel.: +49 (0)631 205 4362; fax: +49 (0)631 205 2496; e-mail:


Despite its relevance for ecology and biodiversity, the stability of spatial microeukaryote diversity patterns in time has received only little attention using gene-based strategies, and there is little knowledge about the relation of spatial vs. temporal variation. We addressed this subject by investigating seasonal fluctuations in protistan communities in three ecologically distinct marine habitats. We analyzed 3360 eukaryote small subunit rRNA gene sequences collected along an O2/H2S gradient in a Norwegian fjord in order to reveal shifts in protistan community composition and structure in three different seasons. In all nine clone libraries, ciliates and stramenopiles accounted for the largest proportion. Yet, as expected, at the phylotype level, the protistan communities from distinct habitats differed significantly, with the number of shared phylotypes between two habitats being as low as 18%. This confirmed previous notions that environmental factors along the stratification gradient shape biodiversity patterns. Surprisingly, the intrahabitat community composition and structure varied at a comparable order of magnitude over time, with only 18–28% phylotypes shared within the same habitat. Our study demonstrates that the consideration of local fluctuations in microeukaryote diversity over time offers additional information for diversity surveys and can significantly contribute to the revelation of spatial protistan community patterns.


The study of biogeography offers insights into mechanisms that generate and maintain diversity (Martiny et al., 2006), and has emerged as an important keystone for understanding microbial diversity, ecology, and evolution (Ramette & Tiedje, 2007). In recent years, the distribution patterns of prokaryotes and microeukaryotes (Pommier et al., 2005; Richards & Bass, 2005) were investigated using molecular tools, and revealed phyloclades that represent microorganisms with restrictions to certain habitats and/or habitat types (Reche et al., 2005; Boenigk et al., 2007; Stoeck et al., 2008), changing our perception of microbial diversity patterns (Martiny et al., 2006). The accompanying recent development of several powerful statistical tools enabled comparison of microbial communities based on sequence information (Schloss et al., 2004; Lozupone & Knight, 2005; Lozupone et al., 2006; Schloss & Handelsman, 2006) and extended the potential of environmental sequencing.

In fact, the analysis of biodiversity patterns benefits from a sound knowledge of the stability of recorded spatial patterns over time (Ramette & Tiedje, 2007), for example the extent of seasonal variation in relation to spatial differences. However, most molecular studies comparing two or more microbial communities with the aim of exploring spatial biodiversity patterns have been based on data acquired from unique sampling events (Green et al., 2004; Pommier et al., 2005; Richards & Bass, 2005; Souza et al., 2006). Especially in eukaryote microbiology, the temporal dimension has been largely neglected, and only a few molecular studies sampled the same site repeatedly (Diez et al., 2001; Massana et al., 2004; Romari & Vaulot, 2004; Medlin et al., 2006; McDonald et al., 2007). However, if temporal processes do indeed play a major role in the formation of biodiversity patterns as suggested by some prokaryotic studies (Yannarell et al., 2003; Yannarell & Triplett, 2005; Newton et al., 2006; Sogin et al., 2006; Ramette & Tiedje, 2007), conclusions about the distribution of microeukaryote diversity arising from sampling events at a single point in time may be biased.

In the present study, we therefore assessed the stability of spatial microeukaryote diversity patterns by analyzing seasonal fluctuations in three ecologically distinct marine habitats. As a model system, we chose protistan communities along an O2/H2S gradient in the Norwegian Framvaren Fjord sampled in three different seasons. An initial study at this site indicated that communities derived from three different layers along the vertical physico-chemical stratification gradient differed remarkably from each other, and were assumed to reflect adaptations to the respective conditions within the anoxic water column (Behnke et al., 2006). For the present investigation, we conducted an additional environmental small subunit rRNA (SSU rRNA) gene inventory, resulting in 3360 sequences in total. We statistically analyzed and compared the phylotype richness of nine clone libraries (three habitats sampled in three different seasons), community membership (incidence of phylotypes) and structure (incidence and abundance), and the taxonomic composition of communities to address the following questions: (1) is there a noteworthy temporal variation within the protistan communities derived from different habitats along the O2/H2S gradient in the Norwegian Framvaren Fjord? (2) If so, are communities derived from the same habitat more similar to each other than communities derived from different depths? (3) And finally, what is the impact of the potential local variation of microeukaryote communities over time on the deduced spatial diversity patterns?

Materials and methods

Sampling site and procedure

We sampled three different protistan communities (depths) along an O2/H2S gradient in the supersulfidic Framvaren Fjord, southwestern Norway (Supporting Information, Fig. S1). With sulfide levels in the bottom water 25 times higher than those in the Black Sea, the fjord contains the highest levels of H2S (up to 8 mM) ever reported for an open anoxic basin (Skei, 1988; Millero, 1991). Our sampling site was located in the central basin of the Framvaren Fjord at 58°09′N, 06°45′E with a water depth of 180 m. Samples for protistan diversity analysis were collected at the oxygen-depleted oxic–anoxic boundary layer (samples IF, oxygen below the detection limit, no sulfide detectable, 18–20 m, depending on the sampling season); the lower redox transition zone, characterized by a high abundance of phototrophic purple sulfur bacteria and first detection of sulfide (sample A1, 21–23 m); and a deeper sulfidic layer, characterized by steep chemical gradients and H2S concentrations between 0.47 and 0.67 mM (sample A2, 35–36 m). In a previous analysis, it was found that these depths harbor significantly different microeukaryote communities (Behnke et al., 2006). The communities were sampled in May 2004 (Behnke et al., 2006, designated as ‘M’ in our clone libraries), November 2004 (N), and September 2005 (S). Samples were taken aboard a small vessel using a TFE-lined, 5-L Niskin bottle (Hydrobios, Kiel). To prevent exposure of the anoxic water to atmospheric oxygen, sampling was performed as described previously (Stoeck et al., 2003).

Analysis of sampling site characteristics

Oxygen concentrations were measured on board using an EOT 196 oxygen probe connected to an Oxi 196 microprocessor (WTW, Germany) together with a Portamess 651-2 microprocessor (Knick, Germany). H2S was determined spectrophotometrically immediately after sampling, based on the methylene blue method (Cline, 1969), using a LASA 10/Plus spectrophotometer (Lange, Germany).

DNA extraction and SSU rRNA gene amplification

Protists were collected on 47-mm Durapore membranes (0.65 μm pore size) as described previously (Stoeck et al., 2003). In order to obtain representative samples, i.e. to compensate for possible vertical variability per depth, we filtered water from up to three different Niskin bottles on one filter. Immediately after filtration (10–15 L per filter), filters were individually frozen (c. −200 °C) in a DNA extraction buffer with Proteinase K (100 μm mL−1 final) (Stoeck & Epstein, 2003). High-molecular-weight DNA was extracted as described in Stoeck et al. (2003). In brief, the samples were heated to 65 °C for 2 h in the extraction buffer. Then, lysates were purified by extraction with an equal volume of chloroform–isoamyl alcohol (24 : 1) and precipitated with a 0.7 volume isopropanol. Potential inhibitors of downstream applications were removed by DNA purification using the resin-based Wizard DNA clean up system (Promega, Madison, WI). The integrity of the total DNA was checked by agarose gel electrophoresis (0.8%). Following the multiple-primer approach (Stoeck et al., 2006), we amplified fragments of the SSU rRNA gene ranging from approximately 1100 bp to nearly full length, using four different eukaryote-specific primer sets as described previously (Behnke et al., 2006). To minimize potential PCR amplification bias, with each individual primer set, we ran three to five separate PCR reactions per sample. Before clone library construction, we pooled the products of all PCR reactions of the same sample that were obtained with the same primer set.

Clone library construction

The PCR products used to construct clone libraries were derived from the oxic–anoxic interface (May 2004: M-IF, November 2004: N-IF, September 2005: S-IF), from the lower redox transition zone (May 2004: M-A1, November 2004: N-A1, September 2005: S-A1), and from a deeper sulfidic layer (May 2004: M-A2, November 2004: N-A2, September 2005: S-A2) using the pGEM-T Vector System cloning kit (Promega). Plasmids were isolated from overnight cultures using the 96-well Directprep kit (Qiagen, Valencia, CA). Three hundred and eighty-four clones per depth and season, evenly distributed among the different primer sets, were partially sequenced (M13F sequencing primer) at MWG-biotech (Ebersheim, Germany) using an Applied Biosystems (ABI) 3730 DNA Stretch Sequencer, with the XL Upgrade and the ABI Prism BigDye Terminator version 3.1 Cycle Sequencing Ready Reaction Kit. Sequence quality assessments, PHRED and PHRAP analysis and assembling were performed using the program CodonCode Aligner v. 1.2.4 (CodonCode Corporation, Dedham, MA). For phylotype grouping (98% and 95% minimal sequence similarity; for reasoning, see Caron et al., 2009), sequences were processed as described previously (Behnke et al., 2006) using dotur (Schloss & Handelsman, 2005). The full-length sequences (at least one representative of each phylotype) of this study have been deposited in the GenBank database under accession numbers DQ310187DQ310369 (Behnke et al., 2006) and EF526713EF527205.

Phylogenetic analyses

Environmental SSU rRNA gene sequences were initially compared with those in GenBank using gapped blast analysis (Altschul et al., 1997) to determine their approximate phylogenetic affiliation. Environmental sequence data, together with their closest GenBank matches, were compiled in ARB (Ludwig et al., 2004) and aligned using the ARB FastAligner utility. Alignments were manually refined using phylogenetically conserved secondary structures. Sequences were added to the ARB tree using the QuickAdd parsimony utility based on conserved and unambiguously aligned positions. Potentially chimeric sequences were identified using secondary structure predictions, the Chimera _Check command version 2.7 provided by the Ribosomal Database Project II (Maidak et al., 2001), and partial treeing analyses (Robison-Cox et al., 1995).

Trees for similarity measurements were constructed using the neighbor-joining (NJ) algorithm implemented in ARB. Furthermore, NJ bootstrap trees were performed using the paup* software package 4.0b10 (Swofford, 2001). Phylogenetic analyses were based on 1214 conserved and unambiguously aligned positions.

Protistan richness estimates

We estimated the total number of phylotypes (operational taxonomic units) in each sampled community using the statistical procedures described in detail elsewhere (Behnke et al., 2006; Hong et al., 2006; Jeon et al., 2006; Zuendorf et al., 2006). In brief, we fit seven candidate parametric abundance models to the observed phylotype frequency counts, selecting a preferred model based on the Pearson χ2 and Akaike information criterion statistics, to obtain a final parametric richness estimate and associated SE. The candidate abundance distributions included the equal-species-size model, and the gamma, lognormal, Pareto, inverse Gaussian, and mixtures of two and of three exponential distributions. The corresponding mixed-Poisson distributions (e.g. the gamma-mixed Poisson) were fitted to the frequency count data (derived from the clone libraries) via maximum likelihood, using custom software on the Velocity Cluster in Cornell's Center for Advanced Computing. The selected parametric model in each case is given in a footnote to Table 1. We also computed various nonparametric statistics using the software spade (Chao & Shen, 2003–2005), particularly the abundance-based coverage estimators (of total richness) ACE and ACE1; either ACE or ACE1 was selected as the preferred nonparametric analysis based on an empirical cut-off value for the coefficient of variation of the frequency count data, as given in the literature (Chao & Shen, 2003–2005). The selected nonparametric estimator in each case is given in a footnote to Table 1. Finally, for multiple comparisons among the nine sampled communities, we controlled the experimentwise (overall) error rate using the Bonferroni correction to the α levels of the tests.

Table 1.   Protistan diversity and richness estimates
Phylotypes estimated
 Parametric modelNonparametric
  • Detected protistan diversity (no. of protistan phylotypes, minimum 98% sequence similarity), and parametric (stochastic abundance model-based) and nonparametric (coverage-based) estimates of number of phylotypes, by library.

  • Inverse Gaussian.

  • ACE.

  • Two-mixed exponential.

  • ACE1.

  • Paired symbols (*, +, #) indicate significant differences (P>0.05) between diversity estimates for the respective libraries.

  • M, May 2004; N, November 2004; S, September 2005; IF, oxic–anoxic interface community; A1, lower redox transition zone community; A2, deep-sulfidic layer deep water community; NA, not available: no parametric model yielded satisfactory fit to data.


Note that although the samples analyzed here are large compared with previous studies, they still do not represent the full richness of the sampled communities. This is displayed graphically by the rarefaction curves (Fig. S2), and numerically by SEs of richness estimates (Table 1), which were on average 25% as large as the estimates themselves. Our statistical analyses properly account for this source of error, but it is irreducible (given the sample) and renders later analyses such as community richness comparisons more challenging.

Jaccard similarity index

spade (Chao & Shen, 2003–2005) was used to calculate the Jaccard index as a measure of similarity between two communities. This index can be calculated based on incidence (Jincidence), abundance (Jabundance), and abundance with adjustment for the effect of unseen shared phylotypes in order to reduce bias due to undersampling (Jadjusted) (Chao et al., 2005, 2006). Analyses were performed as recommended by the authors. Similarity values were transformed into a distance matrix (Marczewski–Steinhaus distances) and used for an unweighted pair group method with arithmetic mean analysis (UPGMA) of the nine unique libraries (Sokal & Michener, 1958).

Community membership

To evaluate community membership initially, we used Venn diagrams and determined set composition. Shared and sample-specific phylotypes were calculated using venny (Oliveros, 2007) and diagrams were constructed using Microsoft® Office PowerPoint®. The sum of characteristics (number of phylotypes) was listed in each section of the diagram. Shared characteristics were listed in the overlapping sections (2- and 3-set phylotypes) and exclusive characteristics were given in the nonoverlapping sections (1-sets).

Community structure

We applied ∫-Libshuff (Schloss et al., 2004), TreeClimber (Schloss & Handelsman, 2006), and UniFrac (Lozupone & Knight, 2005; Lozupone et al., 2006) to compare the community structures of all nine unique libraries. These methods determine whether differences between communities are statistically significant using different criteria to test the null hypothesis that two communities share the same structure. Unlike the adjusted abundance-based Jaccard index, these tests assume that there are no unseen phylotypes in the sample. ∫-Libshuff analyses (default settings, 100 000 iterations) are based on sequence distance matrices. TreeClimber analyses (default settings, 10 000 iterations) use parsimony scores (Fitch, 1971) as criteria to differentiate between two libraries. These scores were applied to phylogenetic ARB (NJ) and paup* (NJ bootstrap) trees. We used UniFrac to perform the UniFrac significance test and an implemented P-test (Martin, 2002). UniFrac analyses were based on phylogenetic trees constructed with ARB. Each of the nine libraries was compared with each other, so that 36 pairwise comparisons were tested. The Bonferroni correction was used in order to control the experimentwise error for the multiple comparisons.


We constructed and compared nine unique eukaryotic clone libraries, originating from three different depths (IF, oxic–anoxic interface; A1, lower redox transition zone; A2, deep sulfidic layer) in the anoxic Framvaren Fjord. Each depth was sampled in three different seasons (M, May 2004, Behnke et al., 2006; N, November 2004; S, September 2005). Out of 3360 clones analyzed, we obtained 2206 protistan sequences; 751 of these are from the sampling event in May 2004 (M-IF=299, M-A1=250, M-A2=202, Behnke et al., 2006), 855 are from November 2004 (N-IF=276, N-A1=322, N-A2=257), and 600 are from September 2005 (S-IF=241, S-A1=99, S-A2=260). The relatively low number of protistan target sequences at the lower redox transition zone in September 2005 is due to a high number of nontarget sequences in the clone library, namely crustacean sequences. Clustering of sequences based on 98% minimum similarity revealed pronounced discrepancies in phylotypes richness between the nine libraries, as values ranged between 18 and 73 different phylotypes (Table 1). Both values are from the deep sulfidic layer (A2), collected in May 2004 (M) and September 2005 (S), respectively. While we carried out all analyses using phylotype grouping at 98% and 95% sequence similarity, the following results and discussion refer to the former only, because the results from both data sets were generally congruent.

Taxonomic diversity

Sequences were widely distributed across the major eukaryote lineages known to date (Table 2). The libraries were dominated by alveolates [accounting for 1500 of 2206 protistan clones (68%) and 107 of 236 phylotypes (45%)], followed by stramenopiles (22% of all phylotypes). Within the alveolates, ciliates were dominant, with 1001 clones (67% of all alveolate clones) and 62 phylotypes (58% of all alveolate phylotypes), followed by dinoflagellates (32% of all alveolate phylotypes), uncultured marine alveolates group I (7%), and perkinsozoa (3%). Stramenopiles were dominated by chrysophytes (29% of all stramenopile phylotypes) and diatoms (25%), followed by uncultured marine stramenopiles (MAST, 16%). Further taxonomic groups that constitute at least 5% of all phylotypes in our sample were choanoflagellates, cryptophytes, and cercozoans. Excavates, discicristates, amoebae, haptophytes, fungi, chlorophytes, and Telonema were less abundant in the clone libraries generated (<5% of all phylotypes).

Table 2.   Higher-level taxonomic distribution of protistan phylotypes
Protistan taxonomyNo. of
No. of clones
  1. The taxonomic arrangement within the table follows the clockwise arrangement of groups depicted in the pie charts (Fig. 1). Phylotypes were defined based on 98% minimum sequence similarity.

Stramenopiles, chrysophytes1564
Stramenopiles, phaeophytes12
Stramenopiles, pelagophytes11
Stramenopiles, dictyophytes337
Stramenopiles, diatoms13107
Stramenopiles, bolidophytes11
Stramenopiles, xanthophytes22
Stramenopiles, bicosoecids22
Stramenopiles, Blastocystis11
Stramenopiles, MAST816
Stramenopiles, oomycetes12
Stramenopiles, labyrinthulids22
Stramenopiles, others11
Alveolates, dinoflagellates34272
Alveolates, perkinsozoa36
Alveolates, uncultured marine alveolates8221
Alveolates, ciliates621001

Richness estimates

Although our SSU rRNA gene inventory comprises 2206 target clones, the number of sequences we analyzed is still insufficient to reveal the protistan richness in this fjord (Fig. S2). We therefore used statistical methods to estimate the protistan phylotype richness in our samples. Parametric estimates ranged between 23.3±4.7 (SE) and 240.8±63.7 phylotypes (Table 1). Both values are from the deep sulfidic layer (A2), collected in May 2004 (M) and September 2005 (S), respectively. Nonparametric estimates were comparable, varying from 21.2±3.2 (M-A2) to 299.4±105.7 (S-A2) (Table 1).

Community comparisons

To explore the extent of spatial and temporal variation of community patterns, we compared protistan communities based on taxonomic composition, community membership (incidence of phylotypes) and structure (incidence and abundance), and phylotype richness.

Taxonomic composition

Generally, the taxon composition of the different communities was very similar on a higher-level taxonomy for all sampling seasons and depths. We identified only a few major eukaryote lineages that were restricted to distinctive communities (Fig. 1), and all of these were detected in very low abundances, i.e. they represented <5% of all protistan phylotypes and clones. Amoebozoa, for example, were detected in one of the seasons only, namely in September 2005 (S-IF and S-A1). Other taxa displayed limitation to specific depths, including fungi, chlorophytes, Excavates, and the genus Telonema. They were reported in at least two seasons, but either in the lower redox transition zone (A1) only (Excavates) or in the lower redox transition zone and below (A1 and A2) (fungi, chlorophytes, and Telonema). A higher taxonomic resolution (phylotype level) revealed a very uneven distribution of taxa across time and locale. For example, the number of ciliate phylotypes varied substantially between the different libraries (minimum=9 in S-A1, maximum=27 in S-A2). Heat-mapping analyses demonstrated that all clone libraries have a distinctively different composition of ciliate phylotypes at the class level as well as at the phylotype level (Fig. 2), and 63% of phylotypes were present in one of the libraries only.

Figure 1.

 Taxonomic affiliation of phylotypes with respect to the nine different SSU rRNA gene sequence sublibraries. Taxonomic affiliation of phylotypes (98% sequence similarity cut-off) detected in three different layers (IF, oxic–anoxic interface; A1, lower redox transition zone; A2, deep sulfidic layer) along the O2/H2S gradient in the Framvaren Fjord in three different seasons (May 2004, November 2004, September 2005).

Figure 2.

 Incidence and abundance of ciliate phylotypes. Heatmaps displaying the incidence and abundance of ciliate phylotypes (98% sequence similarity cut-off) detected in three different layers (IF, oxic–anoxic interface; A1, lower redox transition zone; A2, deep sulfidic layer) along the O2/H2S gradient in the Framvaren Fjord in three different seasons (M, May 2004; N, November 2004; S, September 2005) with respect to assignment of phylotypes to ciliate classes. Columns represent the nine different libraries and rows show the phylotypes. Phylotypes are numbered consecutively per class. Color intensity reflects abundance, with black indicating the highest abundance and light gray indicating minimum abundance, respectively, absence. The data were normalized per phylotype (row, a) and per library (column, b), to provide a better contrast between (a) and within samples (b), respectively. Asterisks indicate the phylotypes detected in one of the nine libraries only. Heatmaps were constructed based on the relative abundances of phylotypes using heatmap builder® version 1.1 (King et al., 2005).

Community membership

Regardless of the sampling season, the proportion of phylotypes that were shared between samples from different depths was low: 22% in May 2004, 34% in November 2004, and 18% in September 2005 (Fig. 3). Unexpectedly, the same applied for a comparison of the taxon composition of the same depth sampled in different seasons (Fig. 4): IF-libraries shared only 28% of phylotypes across the three different sampling seasons. The same proportion of phylotypes was shared among A1-libraries, and only 18% of A2-phylotypes were detected in more than one sampling season.

Figure 3.

 Phylotypes shared between protistan communities derived from different layers within the anoxic water column. Venn diagram showing phylotypes (98% sequence similarity cutoff) shared among SSU rRNA gene sequence sublibraries derived from three sampling depths (IF, oxic–anoxic interface; A1, lower redox transition zone; A2, deep sulfidic layer) along the vertical O2/H2S gradient in the Framvaren Fjord in three different seasons (May 2004, November 2004, September 2005). The area of the rectangles is proportional to the size of the respective clone library. Numbers in the overlapping areas indicate the number of phylotypes shared by relevant libraries (2- and 3-set phylotypes); the overlap area is proportional to amount. The numbers in the nonoverlapping areas correspond to the number of sample-specific phylotypes (1-set phylotypes). The numbers in parentheses indicate the proportion of singletons detected. Percentages show the proportion of 1-set phylotypes per season. Continuous lines show samples derived from the oxic–anoxic interface, dashed lines show samples from the lower redox transition zone, and dotted lines show samples from the deep sulfidic layer.

Figure 4.

 Phylotypes shared between protistan communities sampled in different seasons. Venn diagram showing phylotypes (98% sequence similarity cut-off) shared among SSU rRNA gene sequence sublibraries derived from three sampling events (May 2004, November 2004, and September 2005) at three different depths (IF, oxic–anoxic interface; A1, lower redox transition zone; A2, deep sulfidic layer) along the vertical O2/H2S gradient in the Framvaren Fjord. The area of the rectangles is proportional to the size of the respective clone library. Numbers in the overlapping areas indicate the number of phylotypes shared by relevant libraries (2- and 3-set phylotypes); the overlap area is proportional to amount. The numbers in the nonoverlapping areas correspond to the number of sample-specific phylotypes (1-set phylotypes). The numbers in the parentheses indicate the proportion of singletons detected. Percentages show the proportion of 1-set phylotypes per depth. Continuous lines show samples taken in May 2004, dashed lines show samples November 2004, and dotted lines show samples from September 2005.

While the majority of season-specific taxa were represented by a single clone only (singletons), some season-specific phylotypes reached high relative abundances (Fig. S3). For example, one phylotype that showed high similarity to the ciliate Peritromus kahli (minimum 98.9% sequence identity) was exclusively recovered in November, constituting 156 clones. Other phylotypes that occurred in one of the seasons only included choanoflagellates, haptophytes, chrysophytes, and other ciliates (Fig. S3). Several phylotypes were abundant in two of the three sampling seasons, for example, one cryptophyte phylotype that was missing from the November clone libraries. A choanoflagellate phylotype closely related to Diaphanoeca grandis (minimum 98.9% sequence similarity) was detected with 127 representative sequences in May, one representative in November, and none in September.

Community structure

TreeClimber (Schloss & Handelsman, 2006), Unifrac significance, and the Unifrac P-test (Lozupone & Knight, 2005; Lozupone et al., 2006) were used to test the null hypothesis that community structures are similar. For all possible pairwise comparisons at an experimentwise α level of 0.05, each test rejected the null hypothesis; i.e., they found all pairs of communities to be significantly different. The same applied for ∫-Libshuff analyses, with the exception of the comparisons of N-A1 with N-IF, N-A2, and S-A1.

Jaccard similarity index

We estimated community similarities by calculating the Jaccard index based on incidence (Jincidence), abundance (Jabundance), and abundance with adjustment for the effect of unseen phylotypes (Jadjusted) (Chao et al., 2005; Chao et al., 2006). An UPGMA analysis of Jincidence (Fig. 5a) indicated that all nine protistan communities under study were distinctly different from each other in terms of community composition (presence/absence of phylotypes). In contrast, Jabundance and Jadjusted identified two similarity clusters (Fig. 5b and c). One of these clusters comprises the two interface samples S-IF and M-IF. Interestingly, N-IF clustered with N-A1, N-A2, and M-A1. The protistan communities detected in libraries S-A2, M-A2, and S-A1 could not be assigned to any cluster.

Figure 5.

 UPGMA analysis of pairwise Jaccard indices. UPGMA analysis of pairwise Jaccard indices (98% sequence similarity cut-off) for nine SSU rRNA gene sequence libraries derived from three different sampling seasons (M, May 2004; N, November 2004; S, September 2005) and three different depths (IF, oxic–anoxic interface; A1, lower redox transition zone; A2, deep sulfidic layer) along the vertical O2/H2S gradient in the Framvaren Fjord. Indices correspond to Jaccard indices, calculated based on the incidence of phylotypes (a, Jincidence), based on abundances of phylotypes (b, Jabundance), and based on abundances and adjusted for the effect of unseen shared phylotypes (c, Jadjusted). Ovals indicate similarity clusters.

Phylotype richness comparisons

The parametric richness estimators have asymptotically normal distributions, and this allows us to use approximate two-sample z-tests to make pairwise comparisons of phylotype richness among the nine sampled communities. Thirty-six such comparisons are possible; Bonferroni correction was used to conservatively obtain an experimentwise α level of 0.05 across all 36 tests. The null hypothesis of equal richness was then rejected for the comparisons of M-A2 vs. N-IF, vs. N-A1, and vs. S-A2 (Table 1).


Space and time are thought to be important factors shaping microbial biodiversity patterns (Newton et al., 2006; Sogin et al., 2006; Ramette & Tiedje, 2007). Yet, spatio-temporal variations in protistan plankton communities have hardly been addressed using molecular tools (Diez et al., 2001; Massana et al., 2004; Romari & Vaulot, 2004; Medlin et al., 2006; McDonald et al., 2007). Traditionally, ecological studies on protistan communities in aquatic ecosystems rely on microscopy approaches and the ability to distinguish, identify, and enumerate individual species living in complex microbial assemblages. As noted earlier (Caron et al., 2004), this is a reasonable task for morphologically distinctive species and some higher-taxon rank protistan groups. Because at present, assaying the entire protistan biodiversity in a single microscope-based study is a very difficult task, most, if not all, community studies focus either on a subset of the whole community (Santoferrara & Alder, 2009) or, alternatively, assign protists only to higher taxonomic groups such as ciliates, heterotrophic nanoflagellates (Tanaka & Rassoulzadegan, 2002), or sarcodines (Kling & Boltovskoy, 1995; Michaels et al., 1995), which provide a relatively sound basis for taxon identification and enumeration. Nevertheless, even for ‘easy’ groups (such as the larger sarcodines), immature specimens or species that lack skeletal structures may baffle our ability to properly identify and count these taxa (Dennett et al., 2002). Furthermore, many protists may remain undetected due to their small sizes, while others may remain unrecognized because of their cryptic nature (Nanney et al., 1998; Boenigk et al., 2005; Rodriguez et al., 2005). This situation is complicated when studying extreme environments (such as sulfidic and anoxic waters) because a large proportion of protists remain to be described in such habitats (Foissner, 1999). The ecological study of planktonic protistan communities has thus hardly been extended to extreme aquatic environments.

In order to overcome such difficulties, we applied a microscopy-independent, whole community-targeting molecular strategy (SSU rRNA approach) to assess spatio-temporal patterns of protistan plankton in oxygen-deficient and sulfidic waters in a Norwegian Fjord. Even though the approach described in this study has been applied very successfully for many years to describe protistan communities in a variety of environments (see Epstein & López-Garcia, 2008 for a review), it is certainly not completely reliable either and is prone to biases. For example, in contrast to microscopy-based studies, quantification of taxa based on the number of phylotypes obtained in clone libraries is hardly possible because of large variations in rRNA cistron copies among taxa (Zhu et al., 2005). Abundance-based comparisons may therefore only be legitimate when considering the relative abundance between samples that were analyzed using the same protocol as in this study. Another shortcoming is that some taxa escape detection because of issues such as inadequate nucleic acid extraction (Martin-Laurent et al., 2001) or they are not readily amplifiable due to PCR-primer incompatibilities (Stoeck et al., 2006). Finally, even though Caron et al. (2009) suggested that an SSU rDNA sequence divergence of 2% is a reasonable approximation to discriminate strains of the same species from each other, it is hardly possible to assign phylotypes to a specific taxon (rank). Therefore, the results obtained by high-resolution molecular strategies looking at protistan community members at genotype levels are difficult to compare with traditional plankton ecology studies looking at taxon ranks. Yet, as discussed in the following, the general spatio-temporal patterns we observed in our study mirror the pictures obtained by traditional microscopy-based studies.

Regardless of the sampling season, few protistan phylotypes were shared among the different sampling depths (Fig. 3) and some phylogenetic groups seemed to be restricted to one or two of the depths under study (Fig. 1). This vertical spatial pattern in the water column indicates specific habitat preferences of distinct taxonomic protistan lineages, resulting in a communal division. Our finding is not unexpected as several microscopy-based (Bark & Goodfellow, 1985; Fenchel et al., 1995) as well as molecular-based (Stoeck et al., 2003; Behnke et al., 2006) studies reported distinct community structures along a physico-chemical stratification gradient. Possible explanations for these diversity patterns are several-fold: for example in the oxic–anoxic interface (IF), the growth of photoautotrophs is supported by the availability of photosynthetic-active radiation that is deficient for protistan plankton below the interface (A1 and A2) (Sørensen, 1988). This assumed influence of light availability is supported by the detection of Chl-a (1.22 μg L−1, this study, unpublished data) in the interface. Below the interface, Chl-a was undetectable (using HPLC). Another likely environmental parameter exerting adaptation pressure on the protistan plankton communities along the spatial gradient is H2S increasing from A1 to A2 while being undetectable in the interface. This requires special physiological capabilities and energy metabolisms (Müller, 1993), and may favor the formation of symbiotic consortia (Fenchel & Finlay, 1991). Finally, the quantity and quality of bacterial prey may contribute toward shaping protistan community patterns (Hahn & Höfle, 1999; John & Davidson, 2001; Boenigk et al., 2002). It has been shown that bacterial abundances differ remarkably among the three depths sampled, with the lower redox transition zone displaying the highest bacterial counts (Behnke et al., 2006). These examples mention only a few of the many possible environmental factors shaping protistan plankton communities in the Framvaren Fjord. However, in our study, it was difficult to identify the specific factors influencing the spatial variation in the community composition because the low redundancy of phylotypes in the nine different clone libraries makes multivariate statistical analyses difficult and biased (data not shown).

In addition to spatial variability between different habitats, we observed an unexpected extent of temporal variation within each habitat. Between 72% and 82% of all phylotypes detected per depth were restricted to one of the three sampling seasons (Fig. 4), i.e. temporal variation was as pronounced as spatial differences between depths. Such seasonal fluctuations are not without consequences for the comparison of biodiversity patterns, for example in an ecological context. This becomes evident in the following scenario for the deep sulfidic water depth (A2): the A2-clone library in May was characterized by the lowest number of phylotypes detected and estimated (M-A2, Table 1). However, the same community sampled in September (S-A2) showed the highest diversity compared with all other libraries, and the difference was significant as confirmed by statistical analyses (Table 1). This result challenged previous assumptions that due to the levels of sulfide concentrations (0.47–0.67 mM) and the relatively low bacterial abundances (as a food source for bacterivores), this depth might provide grounds for only a limited number of organisms compared with the oxic–anoxic interface or the lower redox transition zone (Behnke et al., 2006).

Our results demonstrate that it is important to consider the temporal variability of microbial communities, especially when comparing the molecular diversity patterns from different sampling ‘spots.’ This is not surprising considering the wealth of data that have been accumulated from microscopy-based studies assaying spatio-temporal patterns in protistan plankton (Kahru & Nommann, 1990; Kahru et al., 1990; Reul et al., 2005; Rolland et al., 2009). Such studies revealed the control exerted by a variety of biotic and abiotic factors on the spatio-temporal patterns of protistan plankton. Yet, the knowledge gained from these traditional studies has been largely ignored in molecular ecology assays of protistan plankton: several investigations addressed the spatial variation of genetic microbial diversity patterns from distant geographic locations sampled just once. When the samples under comparison were taken at different points in time (Pommier et al., 2005; Richards & Bass, 2005; Souza et al., 2006), this left us with two options for the interpretation of the data: (1) observed differences in diversity patterns either reflect true spatial differences that are due to historical contingencies (geographical distance) or habitat type or (2) these differences result from temporal variation in the local community structures rather than from true spatial differences. In the latter case, communities may be more similar when sampled at corresponding points in time (e.g. the same season), whereas in the first case, community patterns will never be similar regardless of the sampling time.

Seasonal fluctuations in protistan community structures are a consequence of significant temporal population shifts (Bark, 1985; Polat, 2007; Llope et al., 2009). While our data show that the overall composition in terms of major taxonomic groups was generally the same at all sampling dates, the community composition at the phylotype level changed remarkably from season to season within each of the depths investigated. A high number of phylotypes was exclusively present at one of the three sampling seasons (168 taxa) compared with the number present at two or more sampling seasons (68 taxa, Fig. S3). These observations are in line with the idea of a rare biosphere (Sogin et al., 2006) and the seed bank hypothesis (Pedrós-Alió, 2006; Pedrós-Alió, 2007): the microbial biodiversity of any habitat at a given time is characterized by a relatively small number of different populations that dominate all samples, but thousands of low-abundance populations account for the majority of the phylogenetic diversity (Sogin et al., 2006; Huber et al., 2007; Huse et al., 2008). An important property of many microbial taxa is that they can grow and divide without the need for a sexual partner. Thus, they are not barred from becoming abundant by being rare. Rather, it has been hypothesized that low-abundance populations may represent a nearly inexhaustible source of genomic innovation to react to any kind of environmental changes (Pedrós-Alió, 2006; Sogin et al., 2006; Pedrós-Alió, 2007). As a result, previously rare taxa of a microbial community might become dominant in response to environmental changes that favor their growth (Vigil et al., 2009).

In the present study, we detected 127 choanoflagellate clones of the same phylotype in May, but only one in November and none at all in September (Fig. S3, ‘D. grandis’). Provided that PCR amplification of SSU rRNA genes is not entirely random, but dependent on DNA template concentration (Lueders & Friedrich, 2003), this suggests a high relative abundance of the respective organism in May, while it seems to belong to the rare taxa in November and September. As this phylotype seems to be restricted to the deeper sulfidic layer (A2), we might hypothesize that this phylotype is part of a ‘seed bank’ characteristic of the deeper sulfidic layer in the Framvaren Fjord. However, based on the present data, the idea that each of the depths under study harbors its own protistan community with its own specific ‘seed bank’ cannot be verified. To do so, we would need all of the nine libraries to be sampled close to saturation. This would enable us to test whether the ‘seed bank’ memberships of communities derived from the same depth are (1) identical (or at least: very similar) and (2) different from the libraries derived from other depths. However, there is no SSU rRNA gene survey of any community that is reasonably complete; instead, even the largest ones appear to be no more than small samples from apparently endless lists of species present at any locale studied (López-Garcia et al., 2001; Behnke et al., 2006; Stoeck et al., 2006, 2007; Zuendorf et al., 2006; Jeon et al., 2008; Alexander et al., 2009; Edgcomb et al., 2009, Figs S2 and S3). Furthermore, even massively parallel tag sequencing of protistan communities does not seem to sample protistan communities to saturation (Stoeck et al., 2009). Nevertheless, our findings suggest that a large (genetic) pool of biodiversity in each of the depths under study is a basic prerequisite for seasonal fluctuations in protistan communities, presumably as a response to changes in environmental conditions.

In summary, our study demonstrates that in order to reveal the extent and patterns of microeukaryote diversity, environmental inventories substantially benefit from multiple samplings. Temporal variations within communities add an additional dimension to diversity surveys, and therefore, the consideration of local seasonal fluctuations contributes to the revelation of spatial protistan community patterns and the study of biogeography. If resources allow, a combination of microscopy methods combined with a molecular strategy (at least for some selected taxon groups) would provide deeper insights into the ecology and diversity of the investigated protistan communities and would also provide opportunities for correcting the bias of each other.


We thank H.-W. Breiner and J. Maurer for help with sample collection and processing. We acknowledge financial support from the Deutsche Forschungsgemeinschaft to T.S. (STO414/2-2 and STO414/2-3). This research was conducted using the resources of the Cornell University Center for Advanced Computing, which receives funding from Cornell University, New York State, the National Science Foundation, and other leading public agencies, foundations, and corporations; we especially thank Linda Woodard for constructing the database system used for richness estimation and for overseeing the computations. We thank anonymous reviewers for helpful comments on the final version of this manuscript.